Earnings Calls

What this tab is — and the honest disclosure reality

Goldkey Technology Corporation (凌航科技, 3135) does not hold U.S.-style earnings calls, and the corpus contains no company earnings-call transcript. Like most Taiwanese small-caps, Goldkey communicates with investors through periodic investor-conference decks (法人說明會簡報) plus its statutory quarterly financial filings. The company only listed on the TWSE on 6 August 2025, so its entire life as a public company is roughly three quarters long. Two investor communications exist in this corpus and they bookend the story cleanly:

• a pre-IPO roadshow deck dated 15 July 2025 (the "before" picture), and
• a Q2-2026 investor-conference deck (issued in two versions, 29 May and 11 June 2026 — the "after" picture, reporting through 1Q26).

The peer set is far richer, and it is the same industry: four Taiwanese memory-module makers — ADATA (威剛, 3260), TeamGroup (十銓, 4967), Transcend (創見, 2451) and Unifosa (商丞, 8277) — each with several investor-conference decks spanning FY2024→1Q26. These are also decks, not verbatim question-and-answer, but they let us do the one thing a single company snapshot cannot: read the whole industry saying the same thing at the same time.

So this tab does what an earnings-call tab should — extract the live state of the business, the narrative arc, the promise-vs-delivery record and the tells — but it is built from decks and filings, and it says so. Two documents that would have added colour are unusable and I flag them rather than invent around them: Transcend's 116-page Q3_FY2024 file is a corrupted parse (binary garbage, no readable text), and all four 8271 "transcripts" are single blank image pages.

The two-chapter story: commodity module maker → AI-super-cycle earner

Read oldest → newest, Goldkey's own decks tell a stark two-chapter story. In July 2025, on the eve of listing, this was a low-margin memory-module reseller: FY2024 revenue of NT$5,508M at a 4.48% gross margin [1], EPS of NT$2.37 and ROE of 12.32% [2]. The industry section was cautiously constructive — framed as demand "turning better in 2025" driven by AI PCs and rising per-device memory content [3], with supply order "remaining controlled" as the big-three makers pivot capex to HBM [4].

By the Q2-2026 deck, the same business is almost unrecognisable at the margin line: 1Q26 gross margin of 30.4% — a record — on revenue of NT$3,958M, and quarterly EPS of NT$10.76, more than four times the entire FY2024 result [5][6]. Management now frames the memory market as an AI-driven "super-cycle (超級週期)" [7].

Source: Q2-2026 investor deck — 1Q26 gross margin 30.4% and FY2025 revenue NT$7,760M [8]; FY2025 EPS NT$6.33 and 1Q26 EPS NT$10.76 [9]; FY2024 gross margin 4.48% for contrast [10].

The numbers management is putting forward

The revenue line grew steadily for years; the margin line is where the super-cycle shows up. Gross margin went 1.7% (2022) → 5.3% (2023) → 4.5% (2024) → 10.0% (2025) → 30.4% (1Q26) — essentially flat-to-commodity for three years, then a near-vertical step in the most recent quarter.

Source: Q2-2026 investor deck, Financial Highlights (2022–2025 full-year; 2026 = 1Q26) [11].

Source: Q2-2026 investor deck, Financial Highlights [12].

The quarterly revenue trajectory shows the demand pull arriving late — flat/soft through 1H25, then accelerating hard from 3Q25 into a NT$3,958M quarter in 1Q26 (roughly 2.3× the 4Q24 base).

Source: Q2-2026 investor deck, quarterly revenue trend [13].

Earnings compound the effect: EPS was NT$0.30 in 2022 and reached NT$10.76 in a single quarter (1Q26), with FY2025 landing at NT$6.33 (up from NT$2.37 in 2024).

Source: Q2-2026 investor deck, EPS history and key profitability metrics [14].

On capital return, management leans on a long payout record — a 77% average dividend payout over 2015–2024 — and declared the FY2025 distribution as NT$4.1 cash + NT$0.6 stock per share [15]. Note the stock-dividend component and separate commentary about issuing a convertible bond to fund inventory — both are ways to keep cash working through a capital-hungry up-cycle rather than a pure cash-return posture.

What management is now saying — the live narrative

The Q2-2026 deck is unusually explicit about why the margin exploded, and it is entirely a supply-side / AI-demand story rather than a claim of company outperformance:

• Supply is structurally short. The three IDMs are prioritising HBM, capacity is being pulled from commodity parts, and management states plainly: "DDR5 short through 2027, DDR4 capacity squeezed." On the demand side, "CSPs are crowding out non-AI applications; NB / PC / handset brands are broadly raising prices." [16]
• The mechanism is capacity crowd-out. "Capacity shifting to HBM/eSSD → crowds out DDR4/DDR5," with AI servers carrying 8× the DRAM and 3× the NAND of a standard server [17].
• The strategic pivot: de-couple from the price cycle. Management contrasts the "traditional memory-module model" ("low margin, highly dependent on the pricing cycle") with Goldkey's stated core value of raising non-consumer revenue, being anti-cyclical, and generating stable cash flow [18]. The stated 2027 target is to keep lifting the revenue share of high-value industrial / AI / Edge computing [19].
• Where they say the growth is. Management sizes the target end-markets by industry CAGR: robotics/physical-AI at 47.2%, automotive 31.7%, edge-AI/embedded 17.6% (2026–2030) [20], and the reframed product story is a "non-consumer revenue structure" meant to decouple margin from the commodity cycle [21].

What changed between the two decks

Sources: pre-IPO deck industry outlook [22][23]; Q2-2026 deck super-cycle, supply/demand and business-model slides [24][25][26].

Guidance and credibility — reframed for a company that gives none

Goldkey does not publish quantified guidance, so there is no classic guided-vs-delivered scorecard to build. What can be graded is the qualitative call the July-2025 deck made against what the filings then delivered — and on that narrow test, management was directionally right and arguably conservative.

Sources: pre-IPO deck demand/supply framing [27][28]; delivered revenue/margin from Q2-2026 deck [29]; shortage escalation [30].

The credibility caveat that matters: the biggest lift — gross margin from 4.5% to 30.4% — is the cycle, not proof of the anti-cyclical thesis. The whole peer group inflected on the same timeline (next section), so a professional reader should treat Goldkey's "decoupled from the pricing cycle" positioning [31] as an aspiration to be tested on the way down, not an established fact.

Peer / industry cross-read — is everyone saying the same thing?

Yes — emphatically, on the core themes. Every readable peer describes the identical AI-driven memory shortage and shows the same near-vertical margin inflection in the same quarters. This is the single most important finding on the tab: the story is the industry, not the company.

Sources: Goldkey Q2-2026 deck [32]; Transcend 4Q25 gross margin 61.6% [33]; ADATA 2Q25 18.98% [34] and 1Q26 55.69% [35]; TeamGroup 3Q24 11.32% [36], 1Q26 gross profit +140.85% QoQ [37].

Theme-by-theme verdict

Sources: Transcend AI-demand statement [38] and LTA/Fixed-BOM supply lock [39]; ADATA gross margin 48.41% [40]; TeamGroup HBM +88.1% [41]; Goldkey super-cycle [42].

The pricing evidence is the hardest corroboration. Peers publish the contract/spot boards inside their decks, and they all point the same way — a parabolic 2H25→1H26 move:

Sources: ADATA contract DDR4 8G board (to US$85) [43]; TeamGroup contract DDR5 16Gb to US$37.50 [44] and DDR5 spot to US$25.77 [45]. Contract/spot chip prices are quoted in US dollars in the source decks.

The one divergence — the outlier worth naming. Unifosa (8277) is not part of the euphoria. It is a structurally shrunken memory-and-storage distributor that has diversified away from commodity DRAM and remained loss-making even through 2025 — group gross profit did rise +48.07% year-on-year [46] and its small memory segment roughly tripled to NT$35,859K in the first half of 2025 [47], a soft corroboration of the recovery — but it offers no super-cycle language at all. It is the exception that confirms the rule: scale and IDM allocation matter, and the sub-scale distributor gets only a faint version of the up-cycle.

Tone, tells and what to watch

• The loudest tell is that there is no guidance. Goldkey's deck disclaimer states it makes no financial or operating projections. In a super-cycle a management team confident in its own durability often reaches for a number; Goldkey (like its peers) leans on the industry narrative and price boards instead — appropriate humility, but it means the whole bull case rests on the cycle holding.
• Everyone is building inventory into rising prices — including Goldkey. Goldkey is explicitly funding an inventory build with a convertible bond [48]; ADATA pushed inventory to ~38% of assets and TeamGroup to ~62%. This is a leveraged bet on continued price appreciation — a tailwind if prices keep climbing and a sharp reversal risk if they roll.
• The "anti-cyclical" claim is the thing to grade on the way down. Goldkey draws the sharpest contrast — traditional module makers are "low margin, highly dependent on the pricing cycle" while it aims for "anti-cyclical, stable cash flow" [49]. The next down-cycle is the only real test of whether the industrial/AI mix shift actually decouples margins, or whether 1Q26's 30.4% was simply the top.

Money quotes — management in its own words

Bottom line for the reader: across two Goldkey decks and eleven peer decks, the message is one thing said many ways — an AI-driven memory super-cycle is here, prices and margins have inflected vertically, and the constraint is HBM soaking up wafer capacity. Goldkey is a genuine, if mid-pack, beneficiary (1Q26 gross margin 30.4%, EPS NT$10.76). The two open questions the calls do not settle: how long the shortage lasts (management says DDR5 into 2027), and whether Goldkey's industrial/AI mix shift is real "anti-cyclical" decoupling or just good timing into the peak. Watch the inventory build and the first quarter margins stop rising.